AI Platform and Security - Book Structure

This document describes the logical structure of AI Platform and Security, Book 3 of The Transformed Architecture. The book is organised into six parts with 25 numbered chapters, followed by the glossary, two appendices, the references and a note about the author. The chapters build on each other to provide a complete treatment of how large European enterprises and public authorities build, secure and operate AI platforms. Governance, risk, assurance and organisation are treated in the sibling volume AI Governance.

Table of contents

Part I: Foundations (Chapters 1–5)

The book and the series, usage modes, regulatory requirements as engineering constraints, reference architecture and decision records

Part II: Data, Models and DevSecMLOps (Chapters 6–12)

Data architecture, model lifecycle, the DevSecMLOps loop and its risks, foundation models and retrieval, agents and integration

Part III: Infrastructure (Chapters 13–15)

Infrastructure and sovereignty, cloud versus self-hosted AI, and hybrid agentic work

Part IV: Security (Chapters 16–19)

The AI attack surface, security engineering, AI-enabled attackers and agents that exceed their mandate

Part V: Delivery and Operations (Chapters 20–23)

Evaluation, observability, cost and sustainability, and the path from pilot to production

Part VI: Conclusions (Chapters 24–25)

Conclusions for the two organisations and a consolidated analysis

Appendices and Reference

Glossary, Appendix A (code examples), Appendix B (templates and tools), references and about the author

Chapter structure

Part I: Foundations

Chapter	File	Title	Description
1	01_introduction.md	Introduction to AI Platform and Security	Introduces the book, its readers and scope, how it fits The Transformed Architecture, the usage modes and the two organisations that run through it
2	02_llm_usage_modes.md	Usage Modes and the Platform: What Each Mode Demands	A compact recap of the six usage modes of large language models (Mode 0 to Mode 5), with the full taxonomy left to AI Governance, Chapter 2; for each mode, where the organisation's control point sits and what the platform must provide for gateway and routing, identity and delegated authority, isolation and containment, logging and evidence, and data boundaries, with the later chapters that build each
3	03_regulatory_requirements.md	Regulatory Requirements as Engineering Constraints	Condenses the AI Act, the Cyber Resilience Act, NIS 2, DORA and the GDPR into the engineering requirements a platform must meet
4	04_reference_architecture.md	A Reference Architecture for AI in Large Organisations	Layered reference architecture with a cross-cutting control plane
5	05_ai_decision_records.md	AI Decision Records, Model Cards and the Architecture Repository	Extends ADRs to AI, adds model cards and datasheets, and defines the repository as source of truth

Part II: Data, Models and DevSecMLOps

Chapter	File	Title	Description
6	06_data_architecture.md	Data Architecture for AI	Data products, contracts, lineage, feature and vector stores, synthetic data, lawful basis
7	07_model_lifecycle_mlops.md	Model Lifecycle Management: MLOps and LLMOps as Code	Experiment tracking, registries, promotion gates, deployment strategies, LLMOps
8	08_devsecmlops_loop.md	Setting Up an Effective DevSecMLOps Loop	The end-to-end loop for delivering AI systems, with revision management of every artefact as its backbone
9	09_devsecmlops_risks.md	Risks and Vulnerabilities Across the DevSecMLOps Loop	Stage-by-stage risks, documented incidents and controls across the loop, with revision management as the decisive control
10	10_foundation_models_rag.md	Foundation Models, Retrieval and Grounding	Build, buy, or adapt; general-purpose AI obligations; RAG and grounding
11	11_agentic_architectures.md	Agentic Architectures and Orchestration	Agents as components with bounded autonomy, tool use, MCP, approval gates
12	12_integration_patterns.md	Integrating AI into the Enterprise Landscape	AI gateway, synchronous and event-driven integration, contracts for probabilistic services

Part III: Infrastructure

Chapter	File	Title	Description
13	13_infrastructure_sovereignty.md	Infrastructure, Compute and European Digital Sovereignty	Compute, hybrid inference, European cloud, data residency, confidential computing
14	14_cloud_versus_on_premises_ai.md	Cloud AI Services versus On-Premises AI: Different Challenges	How cost, capacity, model change, security, regulation, operations and lock-in differ between cloud AI services and self-hosted AI, and hybrid patterns
15	15_hybrid_agentic_work.md	Hybrid Agentic Work: Self-Hosted and Cloud Models for Sensitive Information	Provider-neutral agentic tools such as OpenCode used with self-hosted models for sensitive information and cloud models for the rest; routing by information class and jurisdiction, enforcement outside the client

Part IV: Security

Chapter	File	Title	Description
16	16_security_for_ai.md	Security for AI Systems	OWASP LLM Top 10, MITRE ATLAS, prompt injection, poisoning, supply chain
17	17_ai_security_engineering.md	AI Security Engineering in Practice	Threat modelling, securing data, retrieval, gateway, agents and supply chain; red teaming, incident response, security evidence
18	18_when_ai_is_the_threat.md	When AI Is the Threat: AI-Enabled Attackers and the Collapsing Exploitation Window	Adversaries who use AI: patch diffing, the shrinking exploitation window, agentic offensive frameworks and deepfake fraud, and the defences against them
19	19_agents_that_exceed_their_mandate.md	When Your Own Agents Exceed Their Mandate: Containment That Does Not Depend on Intent	The organisation's own agents that exceed their mandate: specification gaming, working around sandboxes, misreporting, and containment, detection and response that do not depend on the agent's intent

Part V: Delivery and Operations

Chapter	File	Title	Description
20	20_evaluation_and_testing.md	Evaluation and Testing of AI Systems	The evaluation pyramid, golden datasets, LLM-as-a-judge, red teaming
21	21_observability_and_drift.md	Observability, Monitoring and Drift in Production	Drift, OpenTelemetry, sampled review, incidents, service level objectives
22	22_cost_and_sustainability.md	Cost, Capacity and Sustainability of AI	Token economics, capacity, FinOps, energy and carbon accounting
23	23_pilot_to_production.md	From Pilot to Production: Scaling AI Across the Organisation	The pilot trap, golden paths, readiness reviews, scaling

Part VI: Conclusions

Chapter	File	Title	Description
24	24_conclusions_for_nordvik_and_hansa.md	Nordvik Region and Hansa Mutual: Platform and Security Conclusions	Platform and security conclusions for Nordvik Region and Hansa Mutual, including Hansa's jurisdiction overlays, Polish service centre and UK and Swiss entities
25	25_conclusion.md	Conclusions: Building and Securing AI Platforms	A consolidated analysis of the platform and security argument, with a short forward look

Appendices and Reference

Item	File	Title	Description
Glossary	26_glossary.md	Glossary	AI, architecture, and regulatory vocabulary
Appendix A	27_appendix_a_code_examples.md	Appendix A: Code Examples and Technical Implementations	Complete artefacts from the chapters
Appendix B	appendix_templates_and_tools.md	Appendix B: Templates and Tools	Directory of reusable platform and security templates and tools
References	28_references.md	References and Sources	Canonical bibliography
About	29_about_the_author.md	About the Author	Background and perspective

Canonical ordering

docs/book_index.json defines the chapter ordering and metadata consumed by the automation, and mkdocs.yml carries the same ordering for the website and the book build. The two must agree; tests/test_book_index.py enforces that they do.